Release Overview
FEAIA v1.2.0 is now rolling out to all users. This is the largest update since v1.0.0, touching the core emotion engine, visual layer, and system-level performance.
Release date: April 1, 2026
Supported OS: Windows 10/11 (64-bit)
Installer size: 54MB (12MB smaller than previous release)
How to update: Right-click the tray icon → Check for Updates, or visit the download page
Core Update: Emotion Engine 2.0
The most significant change in v1.2.0 is migrating the emotion recognition engine from rule-based matching to a lightweight neural network architecture.
Technical changes:
- Emotion categories expanded from 6 to 14 (new: curious, fatigued, focused, confident, and more)
- Recognition accuracy on our internal test set: 67% → 94% (40% relative improvement)
- Inference latency: 340ms average → <80ms (effectively real-time)
What this means in practice:
When you're working under deadline pressure, v1.2.0 can now distinguish between "anxious" and "focused" — and adjust accordingly. It won't interrupt your flow with motivational messages when you're deep in concentration; it waits for the right moment instead.
Q: Does emotion sensing data get sent to the cloud?
No. Emotion inference runs entirely on-device regardless of whether you use local or cloud AI mode. No emotional state data ever leaves your machine. This is an explicit design commitment in v1.2.0.
16 New Live2D Skins
The skin library expands from 38 to 54 skins, with new characters across 5 style categories:
| Style Series | New Skins | Featured Characters |
|---|---|---|
| Classical Chinese | 4 | Tea Warmth, Ink Frost |
| Cyberpunk Future | 4 | Quantum Blue, Neon Orange |
| Nature Spirit | 3 | Rain Dew, Forest Voice |
| Dark Night | 3 | Twilight Violet, Stardust Silver |
| Limited Edition | 2 | Spring Festival (available through April 30) |
Q: How do I unlock the new skins?
- Free tier: Access all skins labeled "Free" (currently 12 skins)
- Pro tier: Access ~70% of the library (~38 skins)
- Ultimate tier: Full access to all 54 skins, including all future additions
Performance Improvements
Memory usage:
| Version | Idle | Active Conversation |
|---|---|---|
| v1.1.0 | 180MB | 220MB |
| v1.2.0 | 142MB | 178MB |
A ~21% reduction, achieved through texture compression algorithm upgrades in the Live2D renderer and memory reuse optimizations in the emotion engine.
CPU usage:
- Idle state CPU: reduced from 1.2% to <0.3%
- Peak CPU during response generation: reduced from 28% to 18%
Ollama 0.3.x Compatibility
v1.2.0 is fully compatible with Ollama 0.3.x's updated API protocol, enabling:
- Streaming output: Character responses now render token by token, eliminating the waiting gap before replies appear
- Extended context window: From 4K to 32K tokens (when using qwen2:7b or larger models)
- Hot model switching: Save multiple model configurations in FEAIA settings and switch between them live, without restarting
Bug Fixes
- Fixed: Memory leak occurring after extended runtime (>72 hours)
- Fixed: Tray icon rendering glitch on Windows 11 24H2
- Fixed: Text truncation in UI when system language is set to Traditional Chinese
- Fixed: Race condition between emotion engine and render thread during GPU-accelerated mode
v1.3.0 Preview
v1.3.0 is planned for June 2026 and will include:
- FEAIA IDOL Season 2 debut characters built-in (top 3 unlock free for all users)
- Long-term memory system 2.0: category management and manual memory editing
- Multi-character scene mode (two companions on-screen with synchronized interaction)
- macOS public beta (Intel + Apple Silicon)
Update to v1.2.0 now and experience the upgraded emotion awareness firsthand. Go to Download →
